查看更多>>摘要：Most of the current research on user friendship speculation in location-based social networks is based on the co-occurrence characteristics of users,however,statistics find that co-occurrence is not common among all users;meanwhile,most of the existing work focuses on mining more features to improve the accuracy but ignoring the time complexity in practical applications.On this basis,a friendship inference model named ITSIC is proposed based on the similarity of user interest tracks and joint user location co-occurrence.By utilizing MeanShift clustering algorithm,ITSIC clustered and filtered user check-ins and divided the dataset into interesting,abnormal,and noise check-ins.User interest trajectories were constructed from user interest check-in data,which allows ITSIC to work efficiently even for users without co-occurrences.At the same time,by application of clustering,the single-moment multi-interest trajectory was further proposed,which increased the richness of the meaning of the trajectory moment.The extensive experiments on two real online social network datasets show that ITSIC outperforms existing methods in terms of AUC score and time efficiency compared to existing methods.
查看更多>>摘要：In order to cope with the heavy labor cost challenge of the manual abnormal spectrum classification and improve the effectiveness of existing machine learning schemes for spectral datasets with interference-to-signal ratios,we proposes a semi-supervised classification of abnormal spectrum signals(SSC-ASS),aimed at addressing some of the challenges in abnormal spectrum signal(ASS)classification tasks.A significant advantage of SSC-ASS is that it does not require manual labeling of every abnormal data,but instead achieves high-precision classification of ASSs using only a small number of labeled data.Furthermore,the method can to some extent avoid the introduc-tion of erroneous information resulting from the complex and variable nature of abnormal signals,thereby improving classification accuracy.Specifically,SSC-ASS uses a memory AutoEncoder module to efficiently extract features from abnormal spectrum signals by learning from the reconstruction error.Additionally,SSC-ASS combines convolutional neural network and the K-means using a DeepCluster framework to fully utilize the unlabeled data.Furthermore,SSC-ASS also utilizes pre-training,category mean memory module and replaces pseudo-labels to further improve the classification accuracy of ASSs.And we verify the classification effectiveness of SSC-ASS on synthetic spectrum datasets and real on-air spectrum dataset.
查看更多>>摘要：The robustness of graph neural networks(GNNs)is a critical research topic in deep learning.Many researchers have designed regularization methods to enhance the robustness of neural networks,but there is a lack of theoretical analysis on the principle of robustness.In order to tackle the weakness of current robustness designing methods,this paper gives new insights into how to guarantee the robustness of GNNs.A novel regularization strategy named Lya-Reg is designed to guarantee the robustness of GNNs by Lyapunov theory.Our results give new insights into how regularization can mitigate the various adversarial effects on different graph signals.Extensive experiments on various public datasets demonstrate that the proposed regularization method is more robust than the state-of-the-art methods such as L1-norm,L2-norm,L21-norm,Pro-GNN,PA-GNN and GARNET against various types of graph adversarial attacks.
查看更多>>摘要：Disentangling facial expressions from other disturbing facial attributes in face images is an essential topic for facial expression recognition.Previous methods only care about facial expression disentanglement(FED)itself,ignoring the negative effects of other facial attributes.Due to the annotations on limited facial attributes,it is difficult for existing FED solutions to disentangle all disturbance from the input face.To solve this issue,we propose an expression complementary disentanglement network(ECDNet).ECDNet proposes to finish the FED task during a face reconstruction process,so as to address all facial attributes during disentanglement.Different from traditional re-construction models,ECDNet reconstructs face images by progressively generating and combining facial appearance and matching geometry.It designs the expression incentive(EIE)and expression inhibition(EIN)mechanisms,in-ducing the model to characterize the disentangled expression and complementary parts precisely.Facial geometry and appearance,generated in the reconstructed process,are dealt with to represent facial expressions and complementary parts,respectively.The combination of distinctive reconstruction model,EIE,and EIN mechanisms ensures the com-pleteness and exactness of the FED task.Experimental results on RAF-DB,AffectNet,and CAER-S datasets have proven the effectiveness and superiority of ECDNet.
查看更多>>摘要：Compared with support vector machine,large margin distribution machine(LDM)has better general-ization performance.The central idea of LDM is to maximize the margin mean and minimize the margin variance simultaneously.But the computational complexity of LDM is high.In order to reduce the computational complexity of LDM,a weighted linear loss LDM(WLLDM)is proposed.The framework of WLLDM is built based on LDM and the weighted linear loss.The weighted linear loss is adopted instead of the hinge loss in WLLDM.This modification can transform the quadratic programming problem into a simple linear equation,resulting in lower computational complexity.Thus,WLLDM has the potential to deal with large-scale datasets.The WLLDM is similar in principle to the LDM algorithm,which can optimize the margin distribution and achieve better generalization performance.The WLLDM algorithm is compared with other models by conducting experiments on different datasets.The experimental results show that the proposed WLLDM has better generalization performance and faster training speed.
查看更多>>摘要：Head and neck cancer is one of the most common malignancies in the world.We propose SE-nnU-Net by adapting SE(squeeze and excitation)normalization into nnU-Net,so as to segment head and neck tumors in PET/CT images by combining advantages of SE capturing features of interest regions and nnU-Net configuring itself for a specific task.The basic module referred to convolution-ReLU-SE is designed for SE-nnU-Net.In the encoder it is combined with residual structure while in the decoder without residual structure.The loss function combines Dice loss and Focal loss.The specific data preprocessing and augmentation techniques are developed,and specific network architecture is designed.Moreover,the deep supervised mechanism is introduced to calculate the loss function using the last four layers of the decoder of SE-nnU-Net.This SE-nnU-net is applied to HECKTOR 2020 and HECKTOR 2021 challenges,respectively,using different experimental design.The experimental results show that SE-nnU-Net for HECKTOR 2020 obtained 0.745,0.821,and 0.725 in terms of Dice,Precision,and Recall,respectively,while the SE-nnU-Net for HECKTOR 2021 obtains 0.778 and 3.088 in terms of Dice and median HD95,respectively.This SE-nnU-Net for segmenting head and neck tumors can provide auxiliary opinions for doctors'diagnoses.
查看更多>>摘要：The key to synthesizing the features of electronic medical records(EMR)big data and using them for specific medical purposes,such as mortality and phenotype prediction,is to integrate the individual medical event and the overall multivariate time series feature extraction automatically,as well as to alleviate data imbalance problems.This paper provides a general feature extraction method to reduce manual intervention and automatically process large-scale data.The processing uses two variational auto-encoders(VAEs)to automatically extract individual and global features.It avoids the well-known posterior collapse problem of Transformer VAE through a uniquely designed"proportional and stabilizing"mechanism and forms a unique means to alleviate the data imbalance problem.We conducted experiments using ICU-STAY patients'.data from the MIMIC-Ⅲ database and compared them with the mainstream EMR time series processing methods.The results show that the method extracts visible and comprehen-sive features,alleviates data imbalance problems and improves the accuracy in specific predicting tasks.
查看更多>>摘要：Snoring is a widespreadoccurrence that impacts human sleep quality.It is also one of the earliest symptoms of many sleep disorders.Snoring is accurately detected,making further screening and diagnosis of sleep problems easier.Snoring is frequently ignored because of its underrated and costly detection costs.As a result,this research offered an alternative method for snoring detection based on a long short-term memory based spiking neural network(LSTM-SNN)that is appropriate for large-scale home detection for snoring.We designed acquisition equipment to collect the sleep recordings of 54 subjects and constructed the sleep sound database in the home environment.And Mel frequency cepstral coefficients(MFCCs)were extracted from these sound signals and encoded into spike trains by a threshold encoding approach.They were classified automatically as non-snoring or snoring sounds by our LSTM-SNN model.We used the backpropagation algorithm based on an alternative gradient in the LSTM-SNN to complete the parameter update.The categorization percentage reached an impressive 93.4％,accom-panied by a remarkable 36.9％ reduction in computer power compared to the regular LSTM model.
查看更多>>摘要：As one of the most important railway signaling equipment,railway point machines undertake the major task of ensuring train operation safety.Thus fault diagnosis for railway point machines becomes a hot topic.Considering the advantage of the anti-interference characteristics of vibration signals,this paper proposes an novel intelligent fault diagnosis method for railway point machines based on vibration signals.A feature extraction method combining vari-ational mode decomposition(VMD)and multiscale fluctuation-based dispersion entropy is developed,which is verified a more effective tool for feature selection.Then,a two-stage feature selection method based on Fisher discrimination and ReliefF is proposed,which is validated more powerful than single feature selection methods.Finally,support vector machine is utilized for fault diagnosis.Experiment comparisons show that the proposed method performs best.The diagnosis accuracies of normal-reverse and reverse-normal switching processes reach 100％ and 96.57％ respectively.Especially,it is a try to use new means for fault diagnosis on railway point machines,which can also provide references for similar fields.
查看更多>>摘要：The fault diagnosis of railway point machines(RPMs)has attracted the attention of engineers and researchers.Seldom have studies considered diverse noises along the track.To fulfill this aspect,a multi-time-scale variational mode decomposition(MTSVMD)is proposed in this paper to realize the accurate and robust fault diag-nosis of RPMs under multiple noises.MTSVMD decomposes condition monitoring signals after coarse-grained pro-cessing in varying degrees.In this manner,the information contained in the signal components at multiple time scales can construct a more abundant feature space than at a single scale.In the experimental validation,a random posi-tion,random type,random number,and random length(4R)noise-adding algorithm helps to verify the robustness of the approach.The adequate experimental results demonstrate the superiority of the proposed MTSVMD-based fault diagnosis.
NETL
NSTL
万方数据